Introduction: why catalogue cleaning matters
Product data cleaning isn’t an urgent concern to most merchants…until the carnage is already visible. Slowly but surely, listings have been failing. Filters have stopped working properly and because customers can’t find what they need, complaints are on the rise. So are returns, due to the specification on the page frequently not matching with the product in the box. Internal teams are having to spend too much of their valuable time correcting the same issues across multiple systems. What may look like an ‘untidy’ catalogue is usually symptomatic of a greater operational malaise – with damaging commercial consequences.
Our eBook lays out how to cleanse your catalogue in a structured way. We’ve written it for those businesses who are constantly dealing with fragmented supplier inputs, legacy spreadsheets, inconsistent attributes, and channel pressure.
First of all, the aim isn’t to do a cleaning blitz and tidy up the data once and for all. What should be your goal is to create a repeatable process which enhances product data quality, supports shorter time-to-market, and provides your business with a sustainably clean foundation for using your PIM effectively, syndicating the best possible product information to marketplaces and channels, and making the best use of AI tools for content enrichment.
1. What does “clean” product data actually mean?
Before you start correcting records, we need to have a common definition of what ‘good’ looks like. For practical purposes, clean product data has six qualities:
- Accuracy: the information matches the real product.
- Completeness: required fields are populated for the category and channel.
- Consistency: values, formats, and naming conventions are standardised.
- Timeliness: specifications, certifications, and statuses are current.
- Validity: data follows the business rules for format and allowed values.
- Uniqueness: each product has one governed master record, not several competing versions.
These six principles align with the broader view of product data quality as usability rather than perfection: data is ‘good’ when teams and channels can wholly rely on it without any need for manual rescue work.
2. The six-phase cleaning framework
Moving onto the steps to follow, a highly reliable catalogue cleansing programme uses the following six phases:
- Discovery and assessment
- Profiling and diagnosis
- Standardisation and normalisation
- Enrichment and completion
- Validation and quality assurance
- Governance and continuous improvement
These should be treated as sequential. After all, cleaning the data before you define the rules could well lead to doing the same work twice.
Phase 1: discovery and assessment
Start by gaining knowledge of where your product data lives and how it moves around the organisation today. From our experience across many business clients, it tends to be located (piecemeal or in various versions) all over – across ERP, ecommerce platforms, supplier files, spreadsheets, DAM systems, shared drives, and ad hoc exports. It’s this fragmentation which most often causes a catalogue to deteriorate.
So first, map three things:
- Data sources: every system, spreadsheet, supplier feed, and manual file.
- Stakeholders: product, ecommerce, marketing, operations, compliance, IT.
- Scope: whole catalogue, priority categories, high-value SKUs, or a marketplace subset.
It’s better to do a phased scope than tackling it all at once. Cleaning a priority category thoroughly is of more value than doing the whole catalogue badly.
Phase 2: profiling and diagnosis
Once you’ve clarified where the data is, you can profile its condition. This is the stage where you’re replacing guesswork with evidence!
You need to assess:
- Completeness: which mandatory fields are most often missing?
- Consistency: where do units, formats, or attribute values vary?
- Accuracy: which fields are known to drift or contradict supplier data?
- Duplication: where do multiple records appear to describe the same product?
- Taxonomy quality: are products in the right categories, with the right variant logic?
You could begin practically, using a simple scorecard by category. Count missing hero images, missing dimensions, irregular units, duplicate SKUs, and stale records which haven’t been touched for a set period. This approach gives you a baseline to work from, as well as a prioritised workload as opposed to a vague sense that “this data smells fishy.”
Phase 3: standardising and normalising
If you’re cleaning data without having established a quality standard, it’s like brushing dust under the carpet. Before you edit records, be sure to define the rules for each key attribute:
- Name and meaning
- Allowed values
- Common units of measure
- Obligatory format
- Mandatory or optional attribute status
- Applicability to which category/ies
This is where controlled vocabularies are very useful. For instance, if one team is using “Black,” another uses “blk,” but the supplier sends “matt black,” your filters, search logic, and channel feeds are sure to break down. Minimise free text entries, standardise values first, and then enforce them at ingestion. Exactly the same principle applies to units and formats: choose centimetres or millimetres, not both; choose a single title format and stick to it.
At the same time, try to rationalise product attributes. So, if your catalogue carries “Colour,” “Colour Name,” and “Shade” (with the resulting overlapping meanings), it makes a lot of sense to collapse them into a clearer schema. All this effort is not only important but well worth the time if you are planning and preparing for a PIM implementation – leave it until later and these unclear structures will be more expensive because they’ve already been built into revised workflows and newly-connected integrations.
Phase 4: enrichment and completion
Once you’ve stabilised the underlying structure you can start filling the gaps. Begin with the fields which matter most to customers’ buying decisions, product discoverability (SEO), and compliance/sustainability documentation.
You should prioritise:
- Titles and core identifiers
- Dimensions, weights, and materials
- Compatibility and technical specifications
- Compliance and safety documents
- Images, other visual assets, manuals, and required certification/documentation
- Channel-specific metadata
Not every missing field deserves equal urgency. Price, SKU, title, and essential technical details are clearly critical, but things like meta tags or secondary imagery, while still important, can be addressed later in the sequence. A useful rule of thumb is to enrich those attributes which most directly affect search, filtering, comparison, conversion, and regulatory confidence.
Supplier templates help a lot here. When you can provide structured templates for suppliers, with required fields, approved values, and clear units (and get them to play ball, obviously!) it will improve data quality at source. At Start with Data, we’ve also developed our very handy AI-powered SKULaunch platform. It takes the headaches out of managing all areas of product data onboarding. Getting suppliers normalised as early as possible reduces your downstream workload and reduces time-to-market.
AI tools support enrichment as well, especially for extracting attributes, drafting descriptions, translating, and optimising content. The big BUT is that clean, structured source data must come first. If not, AI will help to reproduce the same weaknesses…at scale.
Phase 5: validation and quality assurance
Cleansing isn’t complete when the record looks ‘better’. It’s done once the record has been verified.
The two layers of validation you can apply:
Automated checks
- Mandatory fields present?
- Allowed values enforced?
- Units and formats correct?
- Duplicate identifiers flagged?
- Variant relationships valid?
- Channel-readiness confirmed?
Human review
- Descriptions accurate and brand-aligned?
- Compatibility data sensible?
- Technical and safety claims verified?
- Visuals correctly linked to the SKU?
- High-value products checked against source documentation?
Human-in-the-loop is essential. Automation is great at finding gaps and non-conforming values. Where it’s weaker is, for instance, judging whether a compatibility note, compliance claim, or category placement is commercially and legally sound.
A best practice here is to validate against a source of truth. For supplier products that may be the current specification sheet or feed. For owned products it could be engineering or manufacturing data. Your objectives are internal consistency and factual accuracy.
Phase 6: governance and continuous improvement
A one-off clean-up is useful up to a point, but a catalogue that stays clean generates genuine value. The risk is that in the absence of a robust governance framework, data will inevitably decline in quality again, whether it’s through rushed onboarding, drift in supplier good practice, shortcuts using manual rework, or unclear and vague ownership and accountability. All the above demonstrates why the product data cleaning process must end with controls as opposed to just corrected records.
At minimum, put these protocols in place:
- Named owners for key attributes and categories
- Approval workflows for changes
- Import rules and validation at source
- Duplicate detection and anomaly alerts
- Monthly or quarterly data audits
- Supplier quality tracking
- Dashboard views for quick checks on completeness, consistency, and freshness
This represents the organisational shift you want – from reactive clean-up to preventative control. The most effective teams working with product data don’t wait until customers complain. They monitor quality continuously and make sure any issues are sorted before publication.
3. Where tools fit in
Manual cleaning has its limits. For smaller catalogues, simple tools might pass muster to standardise values, remove duplicates, and run basic checks. However, as the extent and complexity of your catalogue grows, the business case for PIM and structured onboarding becomes a compelling one.
The PIM platform has fast evolved into the go-to centralised system for workflow, data modelling, rich content authoring, multichannel publishing, supplier onboarding, and AI-supported enrichment. When managing large catalogues or selling across multiple channels, PIM is the most practical and logical execution layer for keeping your cleansed data usable in the long term.
Conclusion: clean data is a capability
Granted, catalogue cleaning is detailed work, pretty granular, but it’s far from being ‘admin for admin’s sake’. Do it meticulously and it significantly improves search, reduces returns, supports channel acceptance, minimises manual rework, and makes using AI and automated tasks far more reliable. Keep in mind that your real goal isn’t a one-off cleaning blitz to admire a tidy product catalogue. You’re aiming for a governed product dataset that is accurate enough to trust, structured enough to scale, and strong enough to support growth.
Next steps
If you’re still holding your product catalogue together through a mix of manual rework, copy-pasting supplier spreadsheets, and a dose of native intuition, get in touch with us today at Start with Data and arrange a discovery call. We’ll advise you on data cleansing, getting your product data management structure fit for use, to equip you properly to compete with the best in the fierce world of digital commerce.